Research map / Point clouds / Segmentation and detection
Point cloud semantic segmentation
Segment point cloud scenes with weak supervision and joint semantic-instance learning.
108 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Segmentation and detection, keyword cluster “semantic segmentation · cloud semantic · instance”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 5 of 6 tracked benchmarks.
32 papers in 2023–24 and 33 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 32.0% to 36.3%.
Open questions the data points to
- S3DIS Area 5 (mIoU): best went from 73.8 (up to 2024) to 74.9 in 2025 — Generative Hard Example Augmentation for Semantic Point Cloud Segmentation (CVPR 2025 · 2 citations)
- S3DIS 6-fold (mIoU): best went from 78.7 (up to 2024) to 79.5 in 2025 — Generative Hard Example Augmentation for Semantic Point Cloud Segmentation (CVPR 2025 · 2 citations)
- S3DIS Area 5 (mAcc): best went from 78.6 (up to 2024) to 82.1 in 2025 — FEAST-Mamba: FEAture and SpaTial Aware Mamba Network with Bidirectional Orthogonal Fusion for Cross-Modal Point Cloud Segmentation (AAAI 2025 · 3 citations)
- S3DIS Area 5 (OA): best went from 91.5 (up to 2024) to 92.5 in 2025 — FEAST-Mamba: FEAture and SpaTial Aware Mamba Network with Bidirectional Orthogonal Fusion for Cross-Modal Point Cloud Segmentation (AAAI 2025 · 3 citations)
- ScanNet validation (mIoU): best went from 76.6 (up to 2024) to 77.8 in 2025 — FEAST-Mamba: FEAture and SpaTial Aware Mamba Network with Bidirectional Orthogonal Fusion for Cross-Modal Point Cloud Segmentation (AAAI 2025 · 3 citations)
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2018 · Large-Scale Point Cloud Semantic Segmentation With Superpoint Graphs (CVPR 2018 · 1,546 citations)
- 2019 · Graph Attention Convolution for Point Cloud Semantic Segmentation (CVPR 2019 · 788 citations)
- 2020 · RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds (CVPR 2020 · 2,138 citations)
- 2021 · SCF-Net: Learning Spatial Contextual Features for Large-Scale Point Cloud Segmentation (CVPR 2021 · 307 citations)
- 2022 · Contrastive Boundary Learning for Point Cloud Segmentation (CVPR 2022 · 202 citations)
- 2023 · SCPNet: Semantic Scene Completion on Point Cloud (CVPR 2023 · 91 citations)
- 2024 · OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024 · 142 citations)
- 2025 · Point Cloud Semantic Segmentation with Sparse and Inhomogeneous Annotations (AAAI 2025 · 18 citations)
- 2026 · Biologically-Inspired Evolutionary Domain Symbiosis for Few-shot and Zero-shot Point Cloud Semantic Segmentation (AAAI 2026 · 1 citation)
Earliest papers here
- Large-Scale Point Cloud Semantic Segmentation With Superpoint Graphs (CVPR 2018 · 1,546 citations)
- Pointwise Convolutional Neural Networks (CVPR 2018 · 652 citations)
- SGPN: Similarity Group Proposal Network for 3D Point Cloud Instance Segmentation (CVPR 2018 · 598 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
S3DIS Area 5 · mIoU (higher is better)
19 papers, 2018–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 58.04 | Large-Scale Point Cloud Semantic Segmentation With Superpoint Graphs (CVPR 2018 · 1,546 citations) |
| 2019 | 62.85 | Graph Attention Convolution for Point Cloud Semantic Segmentation (CVPR 2019 · 788 citations) |
| 2021 | 68.73 | Omni-Supervised Point Cloud Segmentation via Gradual Receptive Field Component Reasoning (CVPR 2021 · 56 citations) |
| 2022 | 69.4 | Contrastive Boundary Learning for Point Cloud Segmentation (CVPR 2022 · 202 citations) |
| 2023 | 72.3 | Unified 3D Segmenter As Prototypical Classifiers (NeurIPS 2023 · 2 citations) |
| 2024 | 73.8 | Multi-View Representation is What You Need for Point-Cloud Pre-Training (ICLR 2024 · 0 citations) |
| 2025 | 74.9 | Generative Hard Example Augmentation for Semantic Point Cloud Segmentation (CVPR 2025 · 2 citations) |
S3DIS 6-fold cross-validation · mIoU (higher is better)
9 papers, 2018–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 62.1 | Large-Scale Point Cloud Semantic Segmentation With Superpoint Graphs (CVPR 2018 · 1,546 citations) |
| 2019 | 68.4 | Point Cloud Oversegmentation With Graph-Structured Deep Metric Learning (CVPR 2019 · 197 citations) |
| 2020 | 70 | RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds (CVPR 2020 · 2,138 citations) |
| 2021 | 71.6 | SCF-Net: Learning Spatial Contextual Features for Large-Scale Point Cloud Segmentation (CVPR 2021 · 307 citations) |
| 2024 | 75 | OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024 · 142 citations) |
S3DIS 6-fold · mIoU (higher is better)
8 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 66.3 | Exploiting Local and Global Structure for Point Cloud Semantic Segmentation with Contextual Point Representations (NeurIPS 2019 · 20 citations) |
| 2021 | 72.2 | Semantic Segmentation for Real Point Cloud Scenes via Bilateral Augmentation and Adaptive Fusion (CVPR 2021 · 269 citations) |
| 2022 | 74.7 | Learning Multi-View Aggregation in the Wild for Large-Scale 3D Semantic Segmentation (CVPR 2022 · 93 citations) |
| 2024 | 78.7 | Improved MLP Point Cloud Processing with High-Dimensional Positional Encoding (AAAI 2024 · 10 citations) |
| 2025 | 79.5 | Generative Hard Example Augmentation for Semantic Point Cloud Segmentation (CVPR 2025 · 2 citations) |
S3DIS Area 5 · mAcc (higher is better)
5 papers, 2020–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 61.4 | JSNet: Joint Instance and Semantic Segmentation of 3D Point Clouds (AAAI 2020 · 129 citations) |
| 2023 | 78.6 | Unified 3D Segmenter As Prototypical Classifiers (NeurIPS 2023 · 2 citations) |
| 2025 | 82.1 | FEAST-Mamba: FEAture and SpaTial Aware Mamba Network with Bidirectional Orthogonal Fusion for Cross-Modal Point Cloud Segmentation (AAAI 2025 · 3 citations) |
S3DIS Area 5 · OA (higher is better)
5 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 88.43 | Exploiting Local and Global Structure for Point Cloud Semantic Segmentation with Contextual Point Representations (NeurIPS 2019 · 20 citations) |
| 2024 | 91.5 | Improved MLP Point Cloud Processing with High-Dimensional Positional Encoding (AAAI 2024 · 10 citations) |
| 2025 | 92.5 | FEAST-Mamba: FEAture and SpaTial Aware Mamba Network with Bidirectional Orthogonal Fusion for Cross-Modal Point Cloud Segmentation (AAAI 2025 · 3 citations) |
ScanNet validation · mIoU (higher is better)
5 papers, 2020–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 43.2 | Multi-Path Region Mining for Weakly Supervised 3D Semantic Segmentation on Point Clouds (CVPR 2020 · 152 citations) |
| 2021 | 63.4 | Boundary-Aware Geometric Encoding for Semantic Segmentation of Point Clouds (AAAI 2021 · 57 citations) |
| 2024 | 76.6 | OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024 · 142 citations) |
| 2025 | 77.8 | FEAST-Mamba: FEAture and SpaTial Aware Mamba Network with Bidirectional Orthogonal Fusion for Cross-Modal Point Cloud Segmentation (AAAI 2025 · 3 citations) |
What has been tried
88 papers here report 1,092 measured changes (a component added, removed or swapped, compared with a baseline): helped 72%, no clear effect 4%, hurt 21%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Use query selection with Hungarian matching | ScanNet validation · mAP50 | 73.9 -> 75.9 | No clear effect | OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024 · 142 citations) |
| replace feature optimization with correlation optimization | S3DIS 1-way 1/5-shot · mIoU | 30.67 -> 39.93; 32.58 -> 42.33 | Helped | Rethinking Few-shot 3D Point Cloud Semantic Segmentation (CVPR 2024 · 48 citations) |
| use 2× point upsampling rather than contrastive or alternative generative pretext tasks | S3DIS Area 5 · mIoU | 43.7; 42.6; 45.1; 49.6; 48.2 | Helped | Less Is More: Label Recommendation for Weakly Supervised Point Cloud Semantic Segmentation (AAAI 2024 · 29 citations) |
| Add the task token | ScanNet V2 · AP50 | 59.2 -> 61.1 | Helped | Frozen CLIP Transformer Is an Efficient Point Cloud Encoder (AAAI 2024 · 25 citations) |
| Add point-level label-first sampling while retaining voxel-level random sampling | S3DIS Area 5, 0.01% uniform · mIoU | 44.7 -> 58.4 | Helped | Point Cloud Semantic Segmentation with Sparse and Inhomogeneous Annotations (AAAI 2025 · 18 citations) |
| add Radial Point Migration (RPM) | ScanNetV2 validation · mAP | 51.6 -> 62.3 | Helped | Spherical Mask: Coarse-to-Fine 3D Point Cloud Instance Segmentation with Spherical Representation (CVPR 2024 · 18 citations) |
| Train ODIN jointly on ScanNet 3D and COCO 2D data | ScanNet instance segmentation · mAP | 47.8 -> 49.1; 1.3% absolute improvement | Helped | ODIN: A Single Model for 2D and 3D Segmentation (CVPR 2024 · 17 citations) |
| Set s=1 and p learnable in HiConv rather than use ABF or RBF | S3DIS 2-way 1-shot · mIoU | 67.50 -> 69.12; 59.28 -> 69.12 | Helped | Taylor Series-Inspired Local Structure Fitting Network for Few-shot Point Cloud Semantic Segmentation (AAAI 2025 · 15 citations) |
| Add the CNF diffusion modeling to the segmentation network | ScanNet · mIoU | 76.6 -> 77.9 | Helped | An End-to-End Robust Point Cloud Semantic Segmentation Network with Single-Step Conditional Diffusion Models (CVPR 2025 · 15 citations) |
| Add PP2S preprocessing | S3DIS Area 5, 0.004% · mIoU | 39.7 -> 53.0 | Helped | Weakly Supervised Point Cloud Semantic Segmentation via Artificial Oracle (CVPR 2024 · 14 citations) |
| add deep features alongside coordinates and colors | ScanNet final instance segmentation · AP | 16.5 -> 18.9 | Helped | FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · 13 citations) |
| Add pseudo-label selection to raw 3D VLM pseudo-label training | ScanNet200 5-shot · HM | 33.28 -> 38.35 | Helped | Generalized Few-shot 3D Point Cloud Segmentation with Vision-Language Model (CVPR 2025 · 12 citations) |
What reviewers will expect
Datasets and baselines named in the 17 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- S3DIS · 11 of 17 papers
- ScanNet · 7 of 17 papers
- SemanticKITTI · 4 of 17 papers
- ScanNet v2 · 4 of 17 papers
- nuScenes · 3 of 17 papers
- ScanNet200 · 3 of 17 papers
- ShapeNetPart · 2 of 17 papers
- ScanObjectNN · 2 of 17 papers
Baselines
- AttMPTI · 4 of 17 papers
- PTv3 · 2 of 17 papers
- PTv3+PPT · 2 of 17 papers
- PTv2 · 2 of 17 papers
- RPVNet · 2 of 17 papers
- SphereFormer · 2 of 17 papers
- PointNeXt · 2 of 17 papers
- DGCNN · 2 of 17 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 65 papers; from OpenAlex and the papers' affiliations).
- Nanyang Technological University · 4 papers
- Shanghai AI Laboratory · 3 papers
- Zhejiang University · 3 papers
- Xiamen University · 3 papers
- East China Normal University · 3 papers
- Tencent · 3 papers
- Peking University · 3 papers
- Chinese Academy of Sciences · 3 papers
- University of Chinese Academy of Sciences · 3 papers
- Halmstad University · 3 papers
Use it
Download BibTeX of 39 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- LiDAR segmentation for driving (47 papers)
- Scene completion and instance segmentation (44 papers)
Elsewhere in the topic
- LiDAR voxel and pillar detectors (56 papers)
- Single-stage point-voxel detection (45 papers)
- Multimodal and open-vocabulary 3D detection (24 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).